COMPARATIVE STUDY OF GABOR TRANSFORM AND DUAL TREE - COMPLEX WAVELET TRANSFORM WITH RESPECT TO ARCHITECTURAL DISTORTION IN MAMMOGRAM
Aditi V. Vedalankar, Ramchandra R. Manthalkar · Journal of Critical Reviews · 2020
The linear or piecewise linear pattern of breast structure may get disturbed due to the development of cancer. Mammography is an economic and widely used modality of screening breast cancer. The architectural distortion (AD) is a type of malignancy wherein spiculations distorts definite structure of breast tissue. The spiculations can be detected by various transforms like Gabor filter, dual tree complex wavelet transform (DT-CWT) etc. that possess an important property of directional sensitivity. This paper presents a system that performs classification of ADs from non-AD mammograms. The mammograms undergo contrast limited adaptive histogram equalization (CLAHE) for enhancement of contrast. The oriental information is retrieved, using Gabor filter and dual tree complex wavelet transform. Haralick features are extracted from Gabor filter processed image and from the coefficients of dual tree complex wavelet transform directional subbands respectively. The feature vector is fed to support vector machine (SVM) classifier for binary classification. The SVM classifier validates if the mammogram has architectural distortion or not. The accuracy, sensitivity and specificity yielded by classifier is 91.66%, 85.71% and 100% respectively in case of Gabor Filter and 90.78%, 85.98% and 98.96% respectively in case of DT-CWT. With the use of DT-CWT, overall time required for execution can be reduced with similar performance. We hypothesize that multi-resolution analysis of a mammogram using DT-CWT gives clear indication of disparity in architectural distortion locale. Furthermore, the system will aid detection of early sign of abnormality.